Research map / Vision-language and multimodal / Multimodal LLMs
GenieBlue: Integrating both Linguistic and Multimodal Capabilities for Large Language Models on Mobile Devices
Xudong Lu, Yinghao Chen, Renshou Wu, Haohao Gao, Xi Chen, Xue Yang, Xiangyu Zhao, Aojun Zhou, Fangyuan Li, Yafei Wen, Xiaoxin Chen, Shuai Ren and 1 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add 2M pure-text samples to fine-tuning data (ablation) | BlueLM-V-3B (BlueLM-3B) · MLLM benchmarks | AVG accuracy: 66.19 -> 66.19 | No clear effect |
| add 2M pure-text samples to fine-tuning data (ablation) | BlueLM-V-3B (BlueLM-3B) · LLM benchmarks | AVG score: 43.78 -> 47.03 | Improved |
| add 2M pure-text samples to fine-tuning data (ablation) | BlueLM-V-3B (BlueLM-3B) · AlignBench | score: 3.87 -> 3.84 | No clear effect |
| add 2M pure-text samples to fine-tuning data (ablation) | BlueLM-V-3B (Qwen2.5-3B) · LLM benchmarks | AVG score: 49.29 -> 51.45 | Improved |
| add 2M pure-text samples to fine-tuning data (ablation) | BlueLM-V-3B (Qwen2.5-3B) · MLLM benchmarks | AVG accuracy: 65.01 -> 64.71 | No clear effect |
| replace LoRA with CogVLM-Skip visual experts (ablation) | BlueLM-3B · MLLM benchmarks | AVG accuracy: 60.78 -> 63.94 | Improved |
| replicate blocks at every quarter (Skip) instead of last quarter (Post) (ablation) | GenieBlue (BlueLM-3B) · MLLM benchmarks | AVG accuracy: 62.10 -> 65.52 | Improved |
| replicate blocks at every quarter (Skip) instead of first quarter (Pre) (ablation) | GenieBlue (Qwen2.5-3B) · MLLM benchmarks | AVG accuracy: 60.46 -> 63.26 | Improved |
| use GenieBlue-Skip instead of CogVLM-Skip | BlueLM-3B · MLLM benchmarks | AVG accuracy: 63.94 -> 65.52 | Improved |
| use GenieBlue-Skip instead of CogVLM-Skip | Qwen2.5-3B · MLLM benchmarks | AVG accuracy: 62.87 -> 63.26 | Improved |
| use non-shared base deployment instead of shared base (ablation) | GenieBlue-Skip (BlueLM-3B) · LLM benchmarks | AVG score: 54.40 -> 60.16 | Improved |
| use non-shared base deployment instead of shared base (ablation) | GenieBlue-Skip (Qwen2.5-3B) · LLM benchmarks | AVG score: 52.57 -> 60.29 | Improved |
Closest papers by meaning
- BlueLM-V-3B: Algorithm and System Co-Design for Multimodal Large Language Models on Mobile Devices (CVPR 2025 · similarity 0.91)
- MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases (ICML 2024 · similarity 0.89)
- On Path to Multimodal Generalist: General-Level and General-Bench (ICML 2025 · similarity 0.87)
- Pangea: A Fully Open Multilingual Multimodal LLM for 39 Languages (ICLR 2025 · similarity 0.87)
- MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models (NeurIPS 2025 · similarity 0.86)
- MM1.5: Methods, Analysis & Insights from Multimodal LLM Fine-tuning (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
